Commit f4140ad0 authored by Ben Walker's avatar Ben Walker
Browse files

nvme: Change the deallocate interface to generic dsm 



Provide a convenience wrapper for general purpose dataset
management commands. The previous wrapper for deallocate
was difficult to use correctly and only for deallocate.

Note that the name is "dataset_management" as opposed to
"data_set_management" to match the NVMe specification.
It's questionable whether "dataset" is valid English, but
it is best to match the specification.

Change-Id: Ifc03d66dbabeabe8146968cf8a09f7ac3446ad68
Signed-off-by: default avatarBen Walker <benjamin.walker@intel.com>
parent d7bbac14

doc/nvme/index.txt

+1 −1
Original line number Diff line number Diff line
@@ -46,7 +46,7 @@ Function | Description 
spdk_nvme_probe()                       | \copybrief spdk_nvme_probe()

spdk_nvme_ns_cmd_read()                 | \copybrief spdk_nvme_ns_cmd_read()

spdk_nvme_ns_cmd_write()                | \copybrief spdk_nvme_ns_cmd_write()

spdk_nvme_ns_cmd_deallocate()           | \copybrief spdk_nvme_ns_cmd_deallocate()

spdk_nvme_ns_cmd_dataset_management()   | \copybrief spdk_nvme_ns_cmd_dataset_management()

spdk_nvme_ns_cmd_flush()                | \copybrief spdk_nvme_ns_cmd_flush()

spdk_nvme_qpair_process_completions()   | \copybrief spdk_nvme_qpair_process_completions()

include/spdk/nvme.h

+22 −12
Original line number Diff line number Diff line
@@ -821,26 +821,36 @@ int spdk_nvme_ns_cmd_read_with_md(struct spdk_nvme_ns *ns, struct spdk_nvme_qpai 
				  uint16_t apptag_mask, uint16_t apptag);



/**

 * \brief Submits a deallocation request to the specified NVMe namespace.

 * \brief Submits a data set management request to the specified NVMe namespace. Data set

 *        management operations are designed to optimize interaction with the block

 *        translation layer inside the device. The most common type of operation is

 *        deallocate, which is often referred to as TRIM or UNMAP.

 *

 * \param ns NVMe namespace to submit the deallocation request

 * \param ns NVMe namespace to submit the DSM request

 * \param type A bit field constructed from \ref enum spdk_nvme_dsm_attribute.

 * \param qpair I/O queue pair to submit the request

 * \param payload virtual address pointer to the list of LBA ranges to

 *                deallocate

 * \param num_ranges number of ranges in the list pointed to by payload; must be

 *                between 1 and \ref SPDK_NVME_DATASET_MANAGEMENT_MAX_RANGES, inclusive.

 * \param ranges An array of \ref spdk_nvme_dsm_range elements describing

 		 the LBAs to operate on.

 * \param num_ranges The number of elements in the ranges array.

 * \param cb_fn callback function to invoke when the I/O is completed

 * \param cb_arg argument to pass to the callback function

 *

 * \return 0 if successfully submitted, ENOMEM if an nvme_request

 *	     structure cannot be allocated for the I/O request

 * \return 0 if successfully submitted, negated POSIX errno values otherwise.

 *

 * The command is submitted to a qpair allocated by spdk_nvme_ctrlr_alloc_io_qpair().

 * The user must ensure that only one thread submits I/O on a given qpair at any given time.

 */

int spdk_nvme_ns_cmd_deallocate(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,

				void *payload, uint16_t num_ranges,

				spdk_nvme_cmd_cb cb_fn, void *cb_arg);

 *

 * This is a convenience wrapper that will automatically allocate and construct the correct

 * data buffers. Therefore, ranges does not need to be allocated from pinned memory and

 * can be placed on the stack. If a higher performance, zero-copy version of DSM is

 * required, simply build and submit a raw command using spdk_nvme_ctrlr_cmd_io_raw().

 */

int spdk_nvme_ns_cmd_dataset_management(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,

					uint32_t type,

					const struct spdk_nvme_dsm_range *ranges,

					uint16_t num_ranges,

					spdk_nvme_cmd_cb cb_fn,

					void *cb_arg);



/**

 * \brief Submits a flush request to the specified NVMe namespace.

include/spdk/nvme_spec.h

+17 −1
Original line number Diff line number Diff line
@@ -446,7 +446,23 @@ SPDK_STATIC_ASSERT(sizeof(struct spdk_nvme_cpl) == 16, "Incorrect size"); 
 * Dataset Management range

 */

struct spdk_nvme_dsm_range {

	uint32_t attributes;

	union {

		struct {

			uint32_t af		: 4; /**< access frequencey */

			uint32_t al		: 2; /**< access latency */

			uint32_t reserved0	: 2;



			uint32_t sr		: 1; /**< sequential read range */

			uint32_t sw		: 1; /**< sequential write range */

			uint32_t wp		: 1; /**< write prepare */

			uint32_t reserved1	: 13;



			uint32_t access_size	: 8; /**< command access size */

		} bits;



		uint32_t raw;

	} attributes;



	uint32_t length;

	uint64_t starting_lba;

};

lib/bdev/nvme/blockdev_nvme.c

+13 −6
Original line number Diff line number Diff line
@@ -90,7 +90,7 @@ struct nvme_io_channel { 


#define NVME_DEFAULT_MAX_UNMAP_BDESC_COUNT	1

struct nvme_blockio {

	struct spdk_nvme_dsm_range dsm_range[NVME_DEFAULT_MAX_UNMAP_BDESC_COUNT];

	int	reserved;

};



enum data_direction {
@@ -679,15 +679,22 @@ blockdev_nvme_unmap(struct nvme_blockdev *nbdev, struct spdk_io_channel *ch, 
{

	struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(ch);

	int rc = 0, i;

	struct spdk_nvme_dsm_range dsm_range[NVME_DEFAULT_MAX_UNMAP_BDESC_COUNT];



	if (bdesc_count > NVME_DEFAULT_MAX_UNMAP_BDESC_COUNT) {

		return -1;

	}



	for (i = 0; i < bdesc_count; i++) {

		bio->dsm_range[i].starting_lba =

			nbdev->lba_start + from_be64(&unmap_d->lba);

		bio->dsm_range[i].length = from_be32(&unmap_d->block_count);

		dsm_range[i].starting_lba = nbdev->lba_start + from_be64(&unmap_d->lba);

		dsm_range[i].length = from_be32(&unmap_d->block_count);

		dsm_range[i].attributes.raw = 0;

		unmap_d++;

	}



	rc = spdk_nvme_ns_cmd_deallocate(nbdev->ns, nvme_ch->qpair, bio->dsm_range, bdesc_count,

	rc = spdk_nvme_ns_cmd_dataset_management(nbdev->ns, nvme_ch->qpair,

			SPDK_NVME_DSM_ATTR_DEALLOCATE,

			dsm_range, bdesc_count,

			queued_done, bio);



	if (rc != 0)

lib/nvme/nvme_ns_cmd.c

+12 −7
Original line number Diff line number Diff line
@@ -396,8 +396,10 @@ spdk_nvme_ns_cmd_write_zeroes(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *q 
}



int

spdk_nvme_ns_cmd_deallocate(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, void *payload,

			    uint16_t num_ranges, spdk_nvme_cmd_cb cb_fn, void *cb_arg)

spdk_nvme_ns_cmd_dataset_management(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,

				    uint32_t type,

				    const struct spdk_nvme_dsm_range *ranges, uint16_t num_ranges,

				    spdk_nvme_cmd_cb cb_fn, void *cb_arg)

{

	struct nvme_request	*req;

	struct spdk_nvme_cmd	*cmd;
@@ -406,9 +408,13 @@ spdk_nvme_ns_cmd_deallocate(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpa 
		return -EINVAL;

	}



	req = nvme_allocate_request_contig(payload,

	if (ranges == NULL) {

		return -EINVAL;

	}



	req = nvme_allocate_request_user_copy((void *)ranges,

					      num_ranges * sizeof(struct spdk_nvme_dsm_range),

					   cb_fn, cb_arg);

					      cb_fn, cb_arg, true);

	if (req == NULL) {

		return -ENOMEM;

	}
@@ -417,9 +423,8 @@ spdk_nvme_ns_cmd_deallocate(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpa 
	cmd->opc = SPDK_NVME_OPC_DATASET_MANAGEMENT;

	cmd->nsid = ns->id;



	/* TODO: create a delete command data structure */

	cmd->cdw10 = num_ranges - 1;

	cmd->cdw11 = SPDK_NVME_DSM_ATTR_DEALLOCATE;

	cmd->cdw11 = type;



	return nvme_qpair_submit_request(qpair, req);

}